Key is an unique key for MxJob to store in maps

Get returns a MxJob

List returns a list of MxJobs

Delete a MxJob

SetPhase set up mxnet job status phase

updateTPRStatus updates the job status based on TraingingJob.status.

Identity will return the locks Identity

Labels returns the labels for this replica set.

Describe is used to convert details on current resource lock into a string

Cleanup cleans up user passed spec, e.g. defaulting, transforming fields. TODO: move this to admission controller

CollectJob collects resources that matches job label, but does not belong to the job with given jobUID

ConfigureAccelerators adds any accelerator specific configuration to the pods.

Create a MxJob

Create attempts to create a LeaderElectionRecord

Get returns the LeaderElectionRecord

IsFailed return true if job status failed

RecordEvent is used to record events

SetDefaults sets any unspecified values to defaults

SetState for mxnet job status

ToSelector converts the labels to a selector matching the labels.

Update will update and existing LeaderElectionRecord

deleteResources deletes the replicas

isRetryableTerminationState returns true if a container terminated in a state that we consider retryable.

replicaStatusFromPodList returns a status from a list of pods for a job.

TODO(jlewi): This is sending a clusterEvent to the channel. I think these are events coming from the cluster code and not k8s events.

setup the training job.

tryAcquireOrRenew tries to acquire a leader lease if it is not already acquired, else it tries to renew the lease if it has already been acquired. Ret

HttpClient returns the http client used.

Copy mxnet job status

Delete deletes the replicas

FullyCollect collects resources that were created before, but does not belong to any current running clusters.

TODO(jlewi): We can probably delete this. func replicaSetStatusToProto(r *MXReplicaSet, status *MXReplicaSetStatus) *tpb.MXReplicaSetStatus { p := &

NewLeadereElector creates a LeaderElector from a LeaderElecitionConfig

Run starts the leader election loop

SetReason for mxnet job status

Update a MxJob.

Validate checks that the MxJobSpec is valid.

Watch MxJobs.

acquire loops calling tryAcquireOrRenew and returns immediately when tryAcquireOrRenew succeeds.

createResources creates all the replicas

TODO(jlewi): This delete function was copied from the etcd-operator. Need to figure out what the right thing to do is. Should we be calling deleteRepl

makeClusterConfig creates the Config object from a cluster initializing it with data from the controller. func (c *Controller) makeClusterConfig() clu

renew loops calling tryAcquireOrRenew and returns immediately when tryAcquireOrRenew fails.

func isSpecEqual(s1, s2 spec.MxJobSpec) bool { // TODO(jlewi): Need to implement this function. return false //if s1.Size != s2.Size || s1.Paused !

stop stops the timer and resets the elapsed duration.

triggerCreatePhase sets the phase to MxJobPhaseCreating additional resource creation happens in TrainingJob.run TODO(jlewi): Need to reconcile this fu

watch creates a go routine, and watches the MX cluster kind resources from the given watch version. It emits events on the resources through the retur

AppendRecoveringCondition for mxnet job status

AppendRemovingDeadMember for mxnet job status

TODO(jlewi): Get rid of the append methods that we don't need AppendScalingDownCondition for mxnet job status

AppendUpgradingCondition for mxnet job status

AsOwner return owner reference

CRDName return crd name

TODO(jlewi): CascadeDeletOptions are part of garbage collection policy. Do we want to use this? See https://kubernetes.io/docs/concepts/workloads/cont

Control set cs ControlPaused = false

Create attempts to create a LeaderElectionRecord annotation

Describe is used to convert details on current resource lock into a string

GetLeader returns the identity of the last observed leader or returns the empty string if no leader has yet been observed.

Status returns the status of the replica set.

returns the Identity of the lock